Additional details on the history of U.S. public housing
Public housing policy differed by city and over time, but some general themes can be identified. Two prominent commonalities were the location of projects within cities and the rapid deterioration of housing quality. Public housing construction frequently propagated racial residential segregation and severely restricted the educational and employment opportunities of tenants. Meanwhile, the declining quality of housing contributed to projects’ reputation as last-resort housing riddled with substandard living conditions and high crime rates.
Location of public housing projects
Following the Housing Act of 1937, local housing authorities and city councils, through “cooperation agreements,” determined the number and placement of public housing projects within cities. However, city councils and mayors commonly rejected sites proposed by housing authorities due to pressure from constituents who did not want public housing in their neighborhood. In 1949, Detroit’s new mayor vetoed eight of 12 proposed public housing sites within his first weeks in office (Sugrue 1996). In 1950, Baltimore City Council rejected sites proposed by the local housing authority after residents spoke out against public housing at a council meeting, and in 1951, the Los Angeles City Council did the same (Hunt 2009). By 1952, Chicago’s housing authority “had surrendered site selection power to city hall and the city council, letting them choose sites without debate” (Hunt 2009, 92). This was a common story for many other cities as well, including Washington, D.C. and Baltimore (Goetz 2013).
The influence of constituent opinion via city council and mayoral authority led to locational choices for public housing that commonly supported and strengthened racial residential segregation. This trend has been well documented in various major cities. For example, Chicago’s housing authority estimated similar costs for locating public housing in white and black neighborhoods and initially proposed sites in both. However, by the late 1940s, “the prospect of active opposition from owner-occupied sites made clearing white neighborhoods a time-consuming and politically perilous task” (Hunt 2009, 43). Opposition to public housing was so strong that in 1971 Alderman Frank Kuta said that he would rather go to jail than have public housing in his ward. In Atlanta, public housing construction was part of a full-scale redevelopment plan that included encouraging black migration to the south and west of downtown (Silver and Moeser 1995). Charlotte also located projects miles from the center of downtown, on the far northwest edge of an established black neighborhood “in accordance with [its] vision of [a] sorted-out city” (Hanchett 1998, 238). Other prominent examples include Detroit (Sugrue 1996), Cincinnati (Fairbanks 1988), Philadelphia (Bauman 1987), Memphis, and Richmond (Hanchett 1998).
These location choices also had important implications for the employment and educational opportunities of public housing residents. In many cities, public housing was placed far from employment opportunities and in places with poor access to public transportation, which made it difficult for tenants to find work. In Detroit, authorities located public housing in the inner city while jobs were moving to the suburbs. Both Charlotte and Atlanta chose to build public housing on the edge of town, where there were minimal employment opportunities (Hanchett 1998; Bayor 1996). In some cities, the construction of densely populated public housing in existing black neighborhoods also led to overcrowded and/or segregated black schools. Schools in black neighborhoods in Atlanta with new public housing surpassed capacity (Bayor 1996). In the 1960s, Darius Swann, a black student in Charlotte, challenged the city’s school system for its role in maintaining segregated schools. Mr. Swann’s lawyer focused on the way that the city used zoning policies, public housing, and urban renewal to promote and maintain residential segregation (Bayor 1996). Despite long waitlists for public housing in major cities, the lack of access to jobs and high-quality schools likely made public housing less desirable to households who could afford private housing.
Changes in public housing quality
The decline in the quality of public housing projects over time also played a role in public housing’s growing reputation as housing of last resort. When public housing was first constructed, it offered a big improvement in housing quality for its tenants. Some tenants moved from slums without electricity, indoor plumbing, or enough beds. Public housing provided all of these amenities (Friedrichs 2011). An early tenant in St. Louis’s Pruitt-Igoe project described her apartment as a “poor man’s penthouse,” and tenants in Chicago’s Alteld Gardens described their new home as “paradise” (Friedrichs 2011, Hunt 2009).
Over time, however, the quality and nature of public housing changed. Low quality construction, as well as lack of maintenance due to inadequate funding, caused many of the buildings to deteriorate quickly. Arbitrary limits on construction costs per unit caused cities to take shortcuts during construction, which eventually led to elevated operating and maintenance costs (Goetz 2013). In 1958, Harrison Salisbury described New York City’s Fort Green public housing project in The New York Times, referring to “the broken windows, the missing light bulbs, the plaster cracking from the walls, the pilfered hardware, the cold drafty corridors, [and] the doors on sagging hinges …” (Salisbury 1958, 75). In St. Louis, the Housing Authority cut corners on the construction of the Pruitt-Igoe project by installing knobs that frequently broke on first use and inadequate window frames that allowed for window panes to be blown in by wind pressure (Meehan 1979). They also failed to properly insulate heated pipes or waterproof basement walls (Meehan 1979). By the early 1970s, the project had deteriorated so severely that the city demolished all 33 high-rises. While much of the decline in the physical structure of public housing was due to inadequate funding, many projects were subject to high volumes of vandalism, including broken light fixtures and windows, graffiti, and urination in elevators and stairwells. The high rates of vandalism and crime found in public housing have been blamed on the site design of projects, which led to a lack of defensible space and increased security and maintenance problems (Goetz 2013; Newman 1972).
Determinants of public housing intensity
Some previous research has studied correlates of public housing intensity. Note that the evidence in this literature is not directly applicable to our paper; this is because our sample includes 56 MSAs, whereas the studies we summarize below each use much larger samples of cities or counties.
Aiken and Alford (1970) investigate the characteristics of cities that adopted public housing, dividing characteristics into political culture, concentration of community power, centralization of formal political structure, community differentiation, and community integration. They proxy for political culture with median family income, the percent of children in private schools, and percent voting Democratic in 1964; for concentration of community power with percent high school graduates and percent of registered voters voting; for political structure with presence of a city-manager form of government, percent nonpartisan elections, and percent of city council elected at large; for community differentiation with age and size of the city; and for community integration with percent unemployed and percent migrant. Aiken and Alford also consider poverty measures, such as the percent of housing dilapidated in 1950, the percent of families with less than $3000 income in 1959, and the percent of the population that is nonwhite. They look at simple correlation coefficients between these controls and three measures of public housing implementation: the presence of any public housing, the number of years that it took a place to build their first public housing, and the amount of public housing per person. These correlations reveal that “older and larger cities, those with lower levels of education and income, fewer managers and officials, higher voting turnout and Democratic voting, a lower degree of in-migration, and higher unemployment levels have participated more in public housing” (p. 858). They also find that cities with higher levels of poverty and dilapidated housing are more likely to adopt public housing.
McDonald (2011) also examines the characteristics of cities that adopted public housing during this period. McDonald considers the characteristics of the population (e.g., population, percent nonwhite) as well as the private housing stock (e.g., age of housing stock, percent of owner-occupied units, percent of units with hot water). When he regresses public housing per capita on these characteristics, he finds that the percentage of low-income families and the percentage of nonwhite residents are positively correlated with public housing, while the percentage of old and owner-occupied housing is negatively correlated with public housing.
Shester (2011) also looks at the characteristics of places that adopted public housing. In her dissertation, she uses a sample of almost all counties in the U.S. and regresses 1970 public housing intensity on a wide variety of 1940 controls and state fixed effects. She finds that percent urban, population density, percent black, percent Democrat, and percent Catholic are all positively correlated with public housing. Percent of the labor force in manufacturing and agriculture, percent of owner-occupied units, median persons per rental unit, and median years of schooling are negatively correlated with public housing. Interestingly, even controlling for 16 county-level characteristics and state fixed effects, the R2 is only 0.4, suggesting that a great deal of unexplained, idiosyncratic variation remains.
Consistent MSA definitions
We limit our sample to individuals who lived in metropolitan statistical areas for which we could define consistent geographic boundaries over time. We used information from IPUMS on individuals’ reported state, county, and metropolitan area in 1950, and information on individuals’ state, county, metropolitan area, and county group in 1970. The 5% 1960 sample does not report county, metropolitan area, or county group, but reports public use micro-data areas (PUMAs). We were able to define consistent boundaries for 56 MSAs in 1950 and 1970 and were able to reconstruct identical boundaries in 1960 for 41 of these MSAs. For the remaining 15, we identified PUMAs that came as close as possible to our 1950 and 1970 consistent boundaries. To assess whether our results were dependent on our choice of 1960 boundaries, we defined two alternative sets of 1960 boundaries for these MSAs. First, we constructed a set of small boundaries, which included only PUMAs that lay completely inside of our consistent boundaries. In this scenario, our small boundaries did not include all of the area in the consistent boundaries but also did not include any area outside of these boundaries. There are three MSAs for which we cannot identify small boundaries. We then constructed a set of large boundaries, which included the entire area in our consistent boundaries as well as additional area in shared PUMAs. In most cases, our preferred boundaries are the small boundaries, as these are usually closer to the consistent ones. However, in some cases the larger ones are closer. For the three MSAs with no small boundaries, the large boundaries are used. In the table below, we list the county composition of our consistent definitions, the census’s MSA definitions in 1950, 1960, and 1970, and, when the 1960 boundaries are slightly different, our preferred, small, and large 1960 boundaries. Our results are virtually unaffected when we use these alternative MSA definitions.
Construction of fertility measure
Before 1970, the census did not ask never-married women about their fertility. To deal with this issue, IPUMS has created a series of rules that allows them to link likely mothers and children within a household. The census reports each individual’s placement in the household (PERNUM) and each individual’s relationship with the head of household (RELATE; e.g., spouse, child, child-in-law, parent, sibling, grandchild, other relative). IPUMS implements a set of rules (MOMRULE) that links women and children. For example, it links individuals listed as children of the head of household to women listed as the household head’s spouse, links the head and siblings of the head to women listed as the head’s mother, and links the head’s wife and siblings-in-law of the head to the head’s mother-in-law. IPUMS also links individuals who are grandchildren of the head of household to a daughter under the following condition:
“Persons listed as the grandchildren are linked to the most proximate preceding (on the form) ever-married daughter, unmarried daughter (if immediately followed by a grandchild), or daughter-in-law of the head, if the daughter/daughter-in-law is 11–59 years older than the grandchild. If no link is formed with a preceding female, the program looks for the most proximate subsequent female who satisfies the same criteria.”Footnote 26
IPUMS uses these rules to create the variable MOMLOC, which identifies the position of an individual’s mother in the household (if applicable), and to create the variable NCHILD, which indicates the number of children each woman has who are living in her household. This NCHILD variable is commonly used to answer questions about single motherhood before 1970 (e.g., Moehling 2007).
We received the new 5% 1960 sample from IPUMS before its official release date, and these imputed variables were not included. In order to make use of the 1960 data, we needed to construct the NCHILD variable for the 1960 data. We implemented the rules described above for our 1950, 1960, and 1970 samples. We compared our measures of mother and never-married single mother for 1950 using IPUMS’ NCHILD definition and our version of it and found remarkably similar results. The table below reports sample means for our mother and never-married single mother variables using both versions of NCHILD for the 1950 and 1970 samples. Our estimates for 1970 are slightly different because IPUMS uses information collected by the census on never-married women’s fertility (CHBORN) in that year, which is unavailable for never-married women in 1950 and 1960.
We prefer our version, as it consistently measures motherhood and single motherhood using the same criteria in all years. Results are similar when using mother and never-married single mother measures defined using IPUMS’ NCHILD variable in 1950 and 1970 and our NCHILD measure for 1960.
Demographic characteristics of public housing residents
Our paper finds effects of public housing on single motherhood rates among black high school dropouts, but not among other race-education groups. Here, we show that it is not surprising that any effect of public housing is only detectable for black dropouts, since they are the group most likely to live in public housing.
We are not aware of data on public housing residents during our study period, but we have used the American Housing Survey (AHS) (United States Bureau of the Census 2016) to characterize the occupants of public housing for 1975–1977, soon after our sample period ends.
There are 22 MSAs that are in both the AHS and our census sample, and the results below are confined to these cities. Our results apply to the population of household heads. We use 1975–1977 because these are the first years in which we can observe both MSA identifiers and the education of the household head in the AHS data. (Education data was not collected for other household members until 1984.)
Table 12 and Fig. 3 show that black high school dropouts are greatly overrepresented in public housing. In Table 12, column 1 is the percentage of black dropouts in the MSA, and column 2 is the percentage of black dropouts among public housing residents in the MSA. The data in these columns are also plotted in Fig. 3. Column 3 is the ratio of the probability of living in public housing among black dropouts in the MSA to the probability of living in public housing among all other race-education groups in the MSA. Public housing is disproportionately composed of black high school dropouts; the median odds ratio is 8.1. Odds ratios for the other three race-education groups are reported in the final three columns of Table 12. Black high school graduates (median odds ratio = 3.4) and nonblack high school dropouts (median odds ratio = 1.5) are slightly overrepresented among public housing residents, but not nearly as severely as black dropouts.
Effects of public housing on single motherhood by age
The results in the main text are for women ages 18–24. As a sensitivity check, we have re-estimated our preferred specification (specification 3) in Tables 3 and 4 for a variety of age ranges. In Tables 13 and 14, we report results from these specifications for ages 19–25, 20–26, 21–27, 22–28, and 23–29, as well as for ages 18–29, and for older women ages 25–34 and 35–44.
In Table 13 (comparable to Table 3), coefficients on public housing for nonblacks and black high school graduates are close to zero for all age bins, as in our main results for ages 18–24 in Table 3. For black high school dropouts, the coefficient is larger for ages 19–25, 20–26, and 21–27, ranging from 0.018 to 0.020, than for our main results for ages 18–24 (0.015). The point estimate is smaller for ages 22–28, 23–29, and 18–29.
The final two columns contain results for older women. The public housing coefficient is 0.0065 for ages 25–34 and 35–44. It is not surprising that the point estimates are smaller for older women because women must be never-married in our sample to be labeled a single mother. The percentage of the sample that has never been married drops sharply with age. For example, for the black 1970 sample, the percentage of never-married women decreases from 54% for ages 18–24, to 24% for ages 23–29, to 9% for ages 35–44.
When we allow for the coefficients to differ by year in Table 14 (comparable to Table 4), the public housing point estimate for black dropouts in 1970 is similar for ages 18–24, 19–25, 20–26, and 21–27, ranging from 0.023 to 0.025. The point estimate again falls for intervals that include older women. Our take on these results is that our findings are not sensitive to the age range that we focus on, but that the effect of public housing on single motherhood, defined as the percentage of women who are mothers and never-married, declines with age.